Human vertebral body slice muscle fat segmentation method, system, electronic device and medium

Abstract

This invention provides a method, system, electronic device, and medium for myolipid segmentation at the vertebral level in the human body. The method includes: acquiring three-dimensional CT medical images of the human vertebral body; resampling the three-dimensional CT medical images to obtain three-dimensional resampled images; generating a semantic segmentation three-dimensional mask covering all regions of the three-dimensional resampled images based on the three-dimensional resampled images; statistically analyzing the pixel percentage of all vertebral body categories included in the semantic segmentation three-dimensional mask along the Z-axis, and generating a one-dimensional dominant vertebral body label sequence based on the pixel percentage of each vertebral body category; obtaining a vertebral level positioning mask based on the one-dimensional dominant vertebral body label sequence using morphological post-processing; and segmenting myolipids within the human vertebral level according to the vertebral level positioning mask. This invention solves the problems of traditional methods relying on end-to-end level annotation and high computational complexity, maintaining sub-millimeter-level positioning accuracy even in low-computing-power scenarios, thus facilitating myolipid segmentation.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical care, to medical image analysis and computer-aided diagnostic technology, and particularly to human vertebral body level localization, specifically a method, system, electronic device and medium for human vertebral body level myofiber segmentation. Background Technology

[0002] Existing vertebral body localization techniques (such as L3 localization) typically rely on high-resolution three-dimensional medical image analysis, requiring comprehensive judgment of large-scale slices along the Z-axis. These techniques have the following drawbacks: Reliance on manual annotation: Existing deep learning models typically employ end-to-end methods, requiring layer annotations, such as L3 start / end slices, for tasks involving layer localization. However, such annotations are lacking in publicly available datasets. Furthermore, existing methods lack an efficient way to convert 3D CT (Computed Tomography) medical image segmentation masks into 1D segment annotations. This limits the use of publicly available datasets for model training, thereby impacting the automation and intelligent development of related diagnoses. Summary of the Invention

[0003] The purpose of this invention is to provide a method, system, electronic device, and medium for myofiber segmentation at the vertebral level of the human body, in order to solve the problems mentioned in the background art.

[0004] In a first aspect, the present invention provides a method for segmenting myofibrils at the vertebral level in the human body. The method includes: acquiring a three-dimensional CT medical image of the human vertebral body; performing resampling processing on the three-dimensional CT medical image to acquire a three-dimensional resampled image; generating a semantic segmentation three-dimensional mask covering all regions of the three-dimensional resampled image based on the three-dimensional resampled image; statistically analyzing the pixel percentage of each vertebral category included in the semantic segmentation three-dimensional mask along the Z-axis, and generating a one-dimensional dominant vertebral label sequence based on the pixel percentage of each vertebral category; obtaining a vertebral level positioning mask based on the one-dimensional dominant vertebral label sequence using morphological post-processing; and segmenting myofibrils within the human vertebral level according to the vertebral level positioning mask.

[0005] In this invention, a three-dimensional segmentation to one-dimensional localization algorithm is designed to analyze the vertebral body distribution layer by layer and obtain the vertebral body level localization mask. Combined with morphological post-processing, the robustness of vertebral body level localization is improved, which solves the problem of traditional methods relying on end-to-end level annotation, and thus facilitates subsequent myolipin segmentation.

[0006] In one implementation of the first aspect, the resampling process of the three-dimensional CT medical image to obtain a three-dimensional resampled image includes: performing voxel resampling in the Z-axis direction of the three-dimensional CT medical image using cubic spline interpolation to standardize the Z-axis slice thickness of the three-dimensional CT medical image, thereby obtaining a standardized image; and performing voxel resampling in the XY plane of the standardized image using linear interpolation to unify the number of pixels in the XY plane matrix of the standardized image, thereby obtaining the three-dimensional resampled image.

[0007] In this implementation, the Z-axis resolution is reduced by XYZ anisotropic voxel resampling, thereby optimizing computational efficiency while preserving the global semantics of the XY plane. This solves the problem of high computational complexity in traditional methods and achieves sub-millimeter-level positioning accuracy even in low-computing-power scenarios.

[0008] In one implementation of the first aspect, the semantic segmentation 3D mask comprises at least one slice along the Z-axis; the step of statistically analyzing the pixel percentage of all vertebrae categories included in the semantic segmentation 3D mask along the Z-axis, slice by slice, to generate a one-dimensional dominant vertebrae label sequence based on the pixel percentage of each vertebrae category includes: for each slice, counting the number of pixels of all vertebrae categories included in the slice; calculating the pixel percentage of all vertebrae categories included in the slice based on the number of pixels; determining whether there is a vertebrae category among all vertebrae categories included in the slice whose pixel percentage exceeds a preset percentage; the preset percentage being greater than or equal to 5. 0%; When the judgment result is yes, the slice is marked as the dominant vertebra; the dominant vertebra is the vertebra category whose pixel percentage exceeds the preset percentage among all vertebra categories contained in the slice; when the judgment result is no, the slice is marked as the background; the one-dimensional dominant vertebra label sequence is generated according to the labeling result; wherein, the one-dimensional dominant vertebra label sequence is a one-dimensional array with a length equal to the number of slices along the Z-axis, each element in the one-dimensional dominant vertebra label sequence corresponds to the background or the dominant vertebra, and the value used to mark the background is different from the value used to mark different dominant vertebrae; the number of slices along the Z-axis is the number of all slices contained in the semantic segmentation three-dimensional mask in the Z-axis direction.

[0009] In this implementation, by converting the semantic segmentation 3D mask into a 1D dominant vertebral body label sequence, it is beneficial to use public datasets for subsequent model training, thereby improving the automation and intelligence of related diagnoses. At the same time, by inputting the 1D dominant vertebral body label sequence into the subsequent model, instead of directly inputting 3D CT medical images into the neural network in the existing technology, the consumption of computing resources and computational complexity are effectively reduced, thus avoiding affecting clinical real-time performance.

[0010] In one implementation of the first aspect, prior to the step of generating a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image, the human vertebral body level myofiber segmentation method further includes: training a 3D segmentation model using a publicly available 3D vertebral body segmentation dataset to obtain a trained 3D segmentation model; the step of generating a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image includes: generating the semantic segmentation 3D mask based on the 3D resampled image using the trained 3D segmentation model.

[0011] In this implementation, by designing a 3D segmentation to 1D localization algorithm and training the 3D segmentation model using a public dataset (i.e., a public 3D cone segmentation dataset), it is possible to utilize the rich semantics of a large public dataset, thus solving the problem that traditional methods limit the use of public datasets for model training.

[0012] In one implementation of the first aspect, the step of obtaining the vertebral level localization mask based on the one-dimensional dominant vertebral label sequence using morphological post-processing includes: performing one-hot encoding on the one-dimensional dominant vertebral label sequence to obtain a first sequence; each first sequence corresponds to a vertebral category, and the number of first sequences is equal to the number of vertebral categories; dilating the first sequence using a dilation operation with a dilation kernel length of a first preset length to fill the mask gaps in the first sequence to obtain a second sequence; eroding the second sequence using an erosion operation with a erosion kernel length of a second preset length to remove isolated noise from the second sequence to obtain a third sequence; determining the length of each vertebral category by traversing the third sequence, and determining whether the length meets the corresponding preset length condition, so that when the length meets the preset length condition and the vertebral category meets the vertebral spatial order constraint rule, the third sequence corresponding to the target category is used as the vertebral level localization mask of the target category; the target category is the vertebral category whose length meets the preset length condition and satisfies the vertebral spatial order constraint rule.

[0013] In this implementation, by performing an optimization operation of first dilation and then erosion on the one-dimensional dominant vertebral label sequence, and combining the vertebral spatial order constraint rules and the length judgment of the vertebral category, the accuracy and reliability of the subsequent vertebral level localization results are ensured.

[0014] In one implementation of the first aspect, segmenting the muscle and fat within the human vertebral body according to the vertebral body level positioning mask includes: extracting a corresponding target region from the three-dimensional CT medical image according to the vertebral body level positioning mask; the target region is the image region in the three-dimensional CT medical image corresponding to the vertebral body category of the vertebral body level positioning mask; and segmenting the muscle and fat within the target region.

[0015] In this implementation, the myofibril segmentation of the human vertebral body is achieved by segmenting the target area segment by segment based on the vertebral body level positioning mask, which is beneficial for subsequent segment-by-segment acquisition of the human vertebral body components.

[0016] In one implementation of the first aspect, after the step of segmenting the muscle and fat within the target region, the human vertebral level muscle and fat segmentation method further includes: obtaining target parameters corresponding to the target region; the target parameters include at least one or a combination of two or more of the following: average HU, total voxel volume, muscle cross-sectional area, and muscle index.

[0017] In this implementation, by obtaining the target parameters corresponding to the target region, the quantitative analysis of myofibrils in each segment of the human vertebral body is achieved.

[0018] Secondly, the present invention provides a human vertebral body-level myofiber segmentation system, the human vertebral body-level myofiber segmentation system comprising: an image acquisition module for acquiring three-dimensional CT medical images of human vertebrae; a resampling module for resampling the three-dimensional CT medical images to acquire three-dimensional resampled images; a mask generation module for generating a semantic segmentation three-dimensional mask covering all regions of the three-dimensional resampled images based on the three-dimensional resampled images; a sequence generation module for statistically analyzing the pixel percentage of all vertebral body categories included in the semantic segmentation three-dimensional mask along the Z-axis, and generating a one-dimensional dominant vertebral body label sequence based on the pixel percentage of each vertebral body category; a processing module for obtaining a vertebral body-level positioning mask based on the one-dimensional dominant vertebral body label sequence using morphological post-processing; and a myofiber segmentation module for segmenting myofiber within the human vertebral body level according to the vertebral body-level positioning mask.

[0019] Thirdly, the present invention provides an electronic device comprising: a processor and a memory; the memory being used to store a computer program; and the processor being used to execute the computer program stored in the memory, so that the electronic device performs the above-described human vertebral body layer myofiber segmentation method.

[0020] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by an electronic device, implements the above-described method for myofiber segmentation at the human vertebral level.

[0021] As described above, the human vertebral body level myofiber segmentation method, system, electronic device, and medium of the present invention have the following beneficial effects: Compared with existing technologies, this invention provides a novel method for achieving precise localization of the human vertebral body without layer annotation. Based on this localization method, myofiber segmentation is achieved, and knowledge from publicly available datasets is reliably incorporated. This reduces the computational complexity of 3D segmentation while maintaining localization accuracy, and effectively suppresses neural network outputs of layer localization and myofiber composition analysis results that do not conform to prior knowledge. Attached Figure Description

[0022] Figure 1 The flowchart shown is a method for myofiber segmentation at the vertebral level according to an embodiment of the present invention.

[0023] Figure 2 The flowchart shown is a process for resampling three-dimensional CT medical images to obtain three-dimensional resampled images, as described in an embodiment of the present invention.

[0024] Figure 3 The flowchart shows the process of calculating the pixel percentage of all vertebrae categories included in the semantic segmentation 3D mask along the Z-axis as described in this embodiment of the invention, and generating a one-dimensional dominant vertebrae label sequence based on the pixel percentage of vertebrae categories.

[0025] Figure 4 The flowchart shown is a process for obtaining a vertebral level localization mask based on morphological post-processing of a one-dimensional dominant vertebral body label sequence, as described in an embodiment of the present invention.

[0026] Figure 5 The flowchart shown is a process for segmenting myolipin within the human vertebral body based on a vertebral body level positioning mask, as described in an embodiment of the present invention.

[0027] Figure 6 The diagram shown is a structural schematic of the human vertebral body level myofiber segmentation system according to an embodiment of the present invention. Detailed Implementation

[0028] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, unless otherwise specified, the following embodiments and features described therein can be combined with each other.

[0029] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0030] See Figures 1 to 6 The following embodiments of the present invention provide a method, system, electronic device, and medium for human vertebral body layer myofiber segmentation. Compared with the prior art, the present invention provides a novel method for achieving accurate localization of human vertebral body layers without layer annotation. Based on this localization method, myofiber segmentation is achieved, and knowledge from publicly available datasets is reliably incorporated. This reduces the computational complexity of 3D segmentation while maintaining localization accuracy, and effectively suppresses neural network outputs of layer localization and myofiber composition analysis results that do not conform to prior knowledge.

[0031] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0032] like Figure 1 As shown, in one embodiment, the present invention provides a method for myofibril segmentation at the vertebral level of the human body, the method comprising: Step S1: Obtain three-dimensional CT medical images of the human vertebrae.

[0033] Step S2: Resample the three-dimensional CT medical image to obtain a three-dimensional resampled image.

[0034] like Figure 2 As shown, in one embodiment, the resampling process of the three-dimensional CT medical image to obtain a three-dimensional resampled image includes: Step S21: Use cubic spline interpolation to resample voxels in the Z-axis direction of the three-dimensional CT medical image to standardize the Z-axis slice thickness of the three-dimensional CT medical image and obtain a standardized image.

[0035] It should be noted that the Z-axis is used as a system axis in 3D CT medical imaging, and it is perpendicular to the "XY plane" mentioned below; the XY plane is the slicing plane. In this invention, a slice refers to a two-dimensional planar image obtained through CT scanning.

[0036] In one embodiment, standardizing the Z-axis slice thickness of the three-dimensional CT medical image includes: standardizing the Z-axis slice thickness of the three-dimensional CT medical image to a preset physical resolution, so as to unify the physical voxel slice thickness.

[0037] It should be noted that the original Z-axis slice thickness (i.e., the spacing in the Z-axis direction) of 3D CT medical images is usually between 0.6 and 5 mm.

[0038] In one embodiment, the preset physical resolution is set to 1.5 mm, that is, the Z-axis slice thickness of the three-dimensional CT medical image is standardized to 1.5 mm.

[0039] Step S22: Perform voxel resampling on the XY plane of the standardized image using linear interpolation to unify the number of pixels in the XY plane matrix of the standardized image, thereby acquiring the three-dimensional resampled image.

[0040] It should be noted that the above-mentioned XYZ anisotropic voxel resampling simulates the image reading mode of a professional physician. It quickly browses and obtains global information in the Z-axis direction, and retains coarse to medium-grained anatomical structures in the XY plane for feature recognition. This design reduces the Z-axis resolution to optimize computational efficiency, while retaining the coarse-grained global semantics of the XY plane. It solves the problem of high computational complexity of traditional methods and achieves sub-millimeter-level positioning accuracy even in low-computing-power scenarios.

[0041] In one embodiment, unifying the number of pixels in the XY plane matrix of the standardized image includes: unifying the number of pixels in the XY plane matrix of the standardized image to 256 pixels × 256 pixels.

[0042] It should be noted that conventional CT scan images themselves show the complete anatomical structure of the human body in the XY axis cross section. For example, a chest CT scan can fully display the anatomical structure of the chest, but does not show the anatomical structure of other Z-axis regions (abdomen, pelvis, etc.). In a typical neural network data processing pipeline, the XY cross section is also segmented, that is, a single neural network inference only processes a part of the data on the cross section (section) (for example, the left chest only contains part of the anatomical structure in the chest XY cross section). In this invention, after using the above-mentioned resampling method, the forward inference of the subsequent 3D segmentation model can always receive the complete cross section anatomical structure, while reducing the computational complexity. Therefore, it retains the global anatomical semantics of the XY cross section and saves computational costs.

[0043] Step S3: Generate a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image.

[0044] In one embodiment, prior to the step of generating a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image, the human vertebral body level myofiber segmentation method further includes: training a 3D segmentation model using a publicly available 3D vertebral body segmentation dataset to obtain a trained 3D segmentation model.

[0045] It should be noted that the publicly available 3D vertebral segmentation dataset uses conventional techniques in the field; for example, in one embodiment, the publicly available 3D vertebral segmentation dataset uses CTSpine1K, which is a large-scale CT dataset specifically for spine segmentation. This dataset aggregates a total of 1005 CT data from four publicly available datasets (excluding slice annotations), and annotates them with 25 vertebral segments: C1-C7, T1-T12, and L1-L6.

[0046] In one embodiment, the three-dimensional segmentation model is SegFormer3D.

[0047] It should be noted that SegFormer3D is a lightweight Transformer architecture that significantly reduces the number of model parameters and computational complexity while maintaining performance.

[0048] Specifically, in practical applications, after the 3D resampled image is input into the 3D segmentation model, the model will output a probability value. By performing corresponding operations on the probability value (which adopts existing technology in the field, so it will not be described in detail here), a semantic segmentation 3D mask can be obtained.

[0049] In one embodiment, the loss function of the 3D segmentation model is a weighted sum of the segmentation loss function and the localization assistance loss function.

[0050] It should be noted that, in the process of training the 3D segmentation model, in order to calculate the localization auxiliary loss function, it is also necessary to convert the semantic segmentation 3D mask into a 1D dominant cone label sequence through the method in step S4 below.

[0051] Among them, the segmentation loss function The calculation formula is as follows:

[0052] It should be noted that this segmentation loss function Used to measure the similarity between a semantic segmentation 3D mask and the actual cone segmentation annotation.

[0053] in, This represents the total number of voxels in the input image (corresponding to the 3D resampled image) input to the 3D segmentation model, and represents the data points of the entire input image; The value ranges from 1 to ; =26 (meaning 26 categories, which include 25 vertebrae covering the first cervical vertebra to the first sacral vertebra and 1 background); The value ranges from 1 to ; Indicates the first Individual elements belong to the first The probability of each category is the probability value output by the 3D segmentation model, representing the confidence of the 3D segmentation model in that each voxel belongs to a specific category. Indicates the first Individual elements belong to the first The true binary labels (0 or 1) for each category are real labeled data obtained from a publicly available 3D cone segmentation dataset, used to train the 3D segmentation model; if the first category... Individual elements do indeed belong to the first Each category, then =1, otherwise =0; It is a constant, which is a very small smoothing term (e.g., 10). -5 ), used to prevent the segmentation loss function The denominator is zero, thus improving the stability of the calculation.

[0054] Location auxiliary loss function The calculation formula is as follows:

[0055] It should be noted that this localization auxiliary loss function Used to assist in vertebral body localization tasks, it regresses the one-dimensional dominant vertebral body label sequence on the Z-axis to ensure that it is as close as possible to the real one-dimensional vertebral body localization label.

[0056] in, This represents the total number of slices in the Z-axis direction of the input image (corresponding to the 3D resampled image) input to the 3D segmentation model; The value ranges from 1 to ; Indicates the first in the Z-axis direction The numerical code of the actual vertebral body category in each slice is obtained by the method in step S4 below; Indicates the first in the Z-axis direction In each slice, the numerical code of the cone category obtained based on the three-dimensional segmentation model (corresponding to the values ​​0, 1, 2, 3, 4, 5 in the following embodiments) is obtained according to the semantic segmentation three-dimensional mask, also through the following step S4; This represents absolute value operations.

[0057] In one embodiment, the loss function of the 3D segmentation model The calculation formula is as follows:

[0058] in, , They represent and The weights are two constants.

[0059] It should be noted that, , The specific values ​​set for these two items are not intended to limit the invention. In practical applications, they can be set according to the specific application scenario, so they will not be described in detail here.

[0060] In this embodiment, generating a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image includes: generating the semantic segmentation 3D mask based on the 3D resampled image using the trained 3D segmentation model.

[0061] Specifically, after obtaining the semantic segmentation 3D mask based on the above 3D segmentation model, it is possible to determine which cone categories are included in the slice, so as to determine the pixel ratio of the cone category included in the slice in the subsequent step S4.

[0062] It should be noted that in this invention, by resampling the three-dimensional CT medical images to obtain three-dimensional resampled images, and then inputting the three-dimensional resampled images into a trained three-dimensional segmentation model, a semantic segmentation three-dimensional mask can be obtained. Compared with the prior art, which directly inputs the full-resolution three-dimensional CT medical images into the neural network, the consumption of computing resources can be reduced, the computational complexity can be reduced, and thus the impact on clinical real-time performance can be avoided.

[0063] Step S4: Along the Z-axis of the semantic segmentation 3D mask, slice by slice, the pixel percentage of all vertebrae categories contained in the semantic segmentation 3D mask is statistically analyzed to generate a one-dimensional dominant vertebrae label sequence based on the pixel percentage of the vertebrae categories.

[0064] like Figure 3 As shown, in one embodiment, the semantic segmentation 3D mask includes at least one slice along the Z-axis; the step of statistically analyzing the pixel percentage of each vertebral category included in the semantic segmentation 3D mask along the Z-axis, slice by slice, to generate a one-dimensional dominant vertebral label sequence based on the pixel percentage of each vertebral category includes: Step S41: For each slice, count the number of pixels of all cone types contained in the slice.

[0065] It should be noted that for any slice, there may be no vertebral category on it, in which case the slice is the background.

[0066] Step S42: Calculate the percentage of pixels of each vertebral body type in the slice based on the number of pixels.

[0067] In this embodiment, the percentage of pixels of a vertebral body category is equal to the number of pixels of that vertebral body category divided by the sum of the number of pixels of all vertebral body categories contained in the slice.

[0068] Step S43: Determine whether there is a vertebral body category whose pixel percentage exceeds a preset percentage among all vertebral body categories included in the slice.

[0069] Specifically, it determines whether there is a vertebral body category whose pixel percentage exceeds a preset percentage among all vertebral body categories contained in the slice, that is, whether the area of ​​a certain vertebra on the slice exceeds the preset percentage of the area of ​​all vertebrae on the slice.

[0070] In this embodiment, the preset percentage is greater than or equal to 50%.

[0071] If the judgment result is yes (that is, it means that among all the vertebral body categories contained in the slice, there is a vertebral body category whose pixel ratio exceeds the preset ratio), then step S44 is executed.

[0072] Step S44: Mark the slice as the dominant vertebral body.

[0073] Specifically, the dominant vertebra is the vertebra category whose pixel percentage exceeds the preset percentage among all vertebra categories included in the slice.

[0074] If the judgment result is negative (that is, it means that there is no vertebral category whose pixel ratio exceeds the preset ratio among all vertebral categories contained in the slice), step S45 is executed.

[0075] Step S45: Mark the slice as background.

[0076] Step S46: Generate the one-dimensional dominant vertebral body label sequence based on the labeling results.

[0077] Specifically, the one-dimensional dominant vertebral label sequence is generated based on the labeling results obtained in steps S44 and S45.

[0078] Wherein, the one-dimensional dominant vertebra label sequence is a one-dimensional array with a length equal to the number of Z-axis slices. Each element in the one-dimensional dominant vertebra label sequence corresponds to the background or the dominant vertebra, and the value used to label the background is different from the value used to label different dominant vertebrae (that is, the labels in steps S44 and S45 are all numerical labels); the number of Z-axis slices is the number of all slices contained in the semantic segmentation three-dimensional mask in the Z-axis direction.

[0079] It should be noted that, considering that in traditional vertebral body localization methods, there is a lack of publicly available end-to-end localization data, which heavily relies on manually labeled layers, and the value of existing public datasets has not been fully explored, this invention designs a 3D to 1D conversion method, which enables the use of information from large public datasets (i.e., public 3D vertebral body segmentation datasets) in the localization task, and can realize the use of the rich semantics of large public datasets.

[0080] Step S5: Based on the one-dimensional dominant vertebral body label sequence, morphological post-processing is used to obtain the vertebral body level localization mask.

[0081] It should be noted that this invention uses the above-mentioned 3D segmentation to 1D localization algorithm to analyze the vertebral body distribution layer by layer and obtain the vertebral body localization mask. Combined with morphological post-processing, it improves the localization robustness and solves the problem of traditional methods relying on end-to-end level annotation.

[0082] like Figure 4 As shown, in one embodiment, the step of obtaining the vertebral body level localization mask based on the one-dimensional dominant vertebral body label sequence using morphological post-processing includes: Step S51: Perform one-hot encoding on the one-dimensional dominant vertebral label sequence to obtain the first sequence.

[0083] In this embodiment, each of the first sequences corresponds to one of the vertebral body categories, and the number of the first sequences is equal to the number of the vertebral body categories.

[0084] It should be noted that in step S51, the one-hot encoding process for the one-dimensional dominant vertebral label sequence will only obtain the first sequence corresponding to the non-background category (i.e., vertebral category). Specifically, this one-hot encoding process means that for each vertebral category, only the value corresponding to that vertebral category is processed as 1, while the values ​​corresponding to the background and other vertebral categories are processed as 0.

[0085] The working principle of step S51 will be further explained below through specific embodiments.

[0086] In one embodiment, the value 0 is used to mark the background, the value 1 is used to mark the vertebral body category T12 (twelfth thoracic vertebra), the value 2 is used to mark the vertebral body category L1 (first lumbar vertebra), the value 3 is used to mark the vertebral body category L2 (second lumbar vertebra), the value 4 is used to mark the vertebral body category L3 (third lumbar vertebra), and the value 5 is used to mark the vertebral body category L4 (fourth lumbar vertebra).

[0087] Specifically, in this embodiment, the one-dimensional dominant cone label sequence obtained through the aforementioned steps S1 to S4 is [0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 4, 4, 5, 5, 0, 0, 0] (that is, the semantic segmentation three-dimensional mask contains 21 slices in the Z-axis direction).

[0088] It should be noted that the one-dimensional dominant vertebral body label sequence has background values ​​at both its head and tail (i.e., the slices at both ends of the semantic segmentation mask in the Z-axis direction are marked as background, that is, in addition to containing the vertebral body category, the three-dimensional CT medical image also contains background at both ends in the Z-axis direction; of course, the three-dimensional CT medical image may only contain the vertebral body category, in which case it is only necessary to supplement the background values ​​at the head and tail of the sequence when generating the one-dimensional dominant vertebral body label sequence). However, the specific number of background values ​​is not a limitation of the present invention. In practical applications, it can be determined according to the length of the expansion kernel in the subsequent step S52 and the length of the erosion kernel in the subsequent step S53.

[0089] After performing one-hot encoding on the one-dimensional dominant vertebral label sequence [0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 4, 4, 4, 4, 4, 5, 5, 0, 0, 0] in step S51, five first sequences are obtained: [0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] and [0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]. The five first sequences [0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0] and [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0] correspond to T12, L1, L2, L3, and L4, respectively.

[0090] Step S52: Dilate the first sequence using a dilation operation with a dilation kernel length of a first preset length to fill the mask gaps in the first sequence and obtain the second sequence.

[0091] In one embodiment, the first preset length is set to 5.

[0092] The working principle of step S52 will be further explained below with reference to the above embodiments.

[0093] In this embodiment, the first sequence is expanded according to an expansion operation with a kernel length of 5.

[0094] Specifically, after obtaining the above five first sequences in step S51, these five first sequences are expanded to obtain five second sequences, namely [0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] and [0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0]. [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0] and [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 0].

[0095] Step S53: Use an etching operation with an etching core length of the second preset length to etch the second sequence to remove isolated noise from the second sequence and obtain the third sequence.

[0096] In one embodiment, the second preset length is set to 5.

[0097] It should be noted that the setting of both the first and second preset lengths to 5 is based on the minimum expected height of the vertebral body; specifically, the expansion operation connects discontinuous regions by taking the maximum value within the window, while the erosion operation eliminates erroneous markers by taking the minimum value within the window.

[0098] It should be noted that morphological post-processing is a conventional technique in the field; in this invention, morphological post-processing includes the expansion operation in step S52 and the etching operation in step S53.

[0099] The working principle of step S52 will be further explained below with reference to the above embodiments.

[0100] In this embodiment, the second sequence is etched according to an etch operation with a core length of 5.

[0101] Specifically, after obtaining the above five second sequences in step S52, each of these five second sequences is eroded to obtain five third sequences, namely [0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] and [0, 0, 0, 0, 0, 0, 1, 1, 1, 0 ... [0], [0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0,0,0], [0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,0,0,0,0,0] and [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0].

[0102] It should be noted that each second sequence obtained in step S52 and each third sequence obtained in step S53 are the same as the first sequence, corresponding to a vertebral body category. The number of second and third sequences is equal to the number of vertebral body categories. That is, the above 5 third sequences correspond to T12, L1, L2, L3 and L4 respectively.

[0103] Step S54: By traversing the third sequence, determine the length of each vertebral body category and determine whether the length meets the corresponding preset length condition. When the length meets the preset length condition and the vertebral body category meets the vertebral body spatial order constraint rule, use the third sequence corresponding to the target category as the vertebral body level positioning mask of the target category.

[0104] Specifically, the target category is the vertebral category whose length satisfies the preset length condition and the vertebral spatial order constraint rule.

[0105] It should be noted that for each vertebral body category, there may be multiple slices, and it is assumed a priori that these slices should be continuous rather than discontinuous.

[0106] In one embodiment, the length of a vertebral body category is equal to the number of the longest consecutive non-zero values ​​in the third sequence corresponding to that vertebral body category multiplied by the slice thickness.

[0107] It should be noted that the slice thickness is set during the resampling process in step S2 above, that is, the standardized Z-axis slice thickness (corresponding to the preset physical resolution in the above embodiment); for the same three-dimensional CT medical image, after the above resampling process, the slice thickness of all corresponding slices is the same.

[0108] In one embodiment, the preset length conditions may be the same or different for different vertebral body types.

[0109] In one embodiment, the preset length condition is that the length is greater than or equal to a preset length threshold.

[0110] Specifically, for different vertebral body types, the preset length thresholds in the corresponding preset length conditions may be the same or different.

[0111] It should be noted that a preset length threshold is set based on the vertebral body anatomical characteristics to eliminate short intervals that do not meet the requirements. This preset length threshold takes into account both physical size (12mm) and the number of slices (approximately 6 layers). For the retained candidate intervals, the longest continuous interval is selected as the corresponding vertebral body localization result.

[0112] In one embodiment, the preset length threshold is set to 10 mm.

[0113] It should be noted that the vertebral spatial sequence constraint rules are pre-defined, and their core purpose is: 1. Mandatory anatomical sequence: Ensure that the four key vertebrae T12, L1, L2, and L3 are arranged strictly in their physiological order along the Z-axis (from top to bottom), that is, T12 must be above L1, L1 above L2, and L2 above L3.

[0114] 2. Handling missing vertebrae: To a certain extent, inferring and filling in vertebrae missing due to incomplete segmentation or noise.

[0115] The working principle of step S54 will be further explained below with reference to the above embodiments.

[0116] In this embodiment, the preset length condition for each vertebral body category is a length greater than or equal to 10 mm, and the slice thickness for each vertebral body category is 4 mm.

[0117] Specifically, after obtaining five third sequences in step S53, by traversing these five third sequences, the lengths of T12, L1, L2, L3, and L4 can be determined to be 3×4=12mm (>10mm), 3×4=12mm (>10mm), 2×4=8mm (<10mm), 5×4=20mm (>10mm), and 2×4=8mm (<10mm), respectively. Then, after judgment, it is found that T12, L1, L2, L3, and L4 conform to the vertebral spatial order constraint rules, but among them, L2 and... The length of L4 does not meet the preset length condition. In this case, the third sequences [0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], [0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], and [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0] corresponding to T12, L1, and L3 are used as the corresponding vertebral body level positioning masks.

[0118] It should be noted that the vertebral body level positioning mask obtained through the above steps represents an actual interval along the long axis of the human vertebral body in the real world; through this vertebral body level positioning mask, the corresponding vertebral body category region image can be extracted from the three-dimensional CT medical image in step S1.

[0119] In one embodiment, when the length does not meet the preset length condition and / or the vertebral body category does not meet the vertebral body spatial order constraint rule, step S5 further includes the following step: issuing an alarm indication.

[0120] In one embodiment, the alarm indication is used to indicate vertebral body categories whose length does not meet a preset length condition, and / or to indicate vertebral body categories that do not meet the vertebral body spatial order constraint rules.

[0121] It should be noted that, considering the uncertainty of neural networks, existing regression-based data-driven methods cannot explicitly guarantee that the output conforms to the prediction results of prior knowledge. That is, the level in which a single vertebra is located should be continuous (one-dimensional simply connected domain). This invention effectively solves the problem in existing vertebral level localization technology that the uncertainty of neural networks leads to the inability to output prediction results that conform to prior knowledge. At the same time, by issuing alarm indications and combining them with the above-mentioned morphological post-processing, this invention also improves the reliability of human-computer interaction in clinical scenarios.

[0122] Step S6: Segment the myofibrils within the human vertebral body according to the vertebral body level positioning mask.

[0123] It should be noted that reusing the vertebral body level positioning mask obtained by vertebral body positioning to achieve myolipin segmentation helps to improve the accuracy of myolipin segmentation, thereby further improving the accuracy of subsequent myolipin quantitative analysis results.

[0124] like Figure 5 As shown, in one embodiment, segmenting the myofibrils within the human vertebral body layer according to the vertebral body layer positioning mask includes: Step S61: Extract the corresponding target region from the three-dimensional CT medical image according to the vertebral body level positioning mask.

[0125] Specifically, the vertebral body level positioning mask is accurately mapped back to the coordinate system of the 3D CT medical image. This process calculates the size ratio between the original space (corresponding to the 3D CT medical image) and the resampling space (corresponding to the vertebral body level positioning mask), and uses an interpolation method to achieve accurate positioning. The final output positioning result includes the start and end slice numbers of the vertebral body in the 3D CT medical image, as well as the corresponding slice world coordinate information. Among them, the slice world coordinate information can be further used for more clinical analysis needs.

[0126] In this embodiment, the target region is the image region in the three-dimensional CT medical image corresponding to the vertebral body category of the vertebral body level positioning mask.

[0127] In one embodiment, the image region is the complete region corresponding to the vertebral body category in a three-dimensional CT medical image, which is formed by extending upward and downward along the Z-axis by a third preset length.

[0128] In one embodiment, the third preset length is set to 5 mm.

[0129] It should be noted that the above design of the imaging region takes into account the anatomical variations of patients with different body types, ensuring that the target region contains complete muscle and fat distribution characteristics. During the extraction of the target region, the spatial coordinate system of the original CT data is strictly maintained to avoid geometric deformation.

[0130] Step S62: Segment the muscle and fat within the target area.

[0131] In one embodiment, segmenting the muscle and fat within the target region includes: segmenting the muscle and fat within the target region using a muscle-fat segmentation network.

[0132] In one embodiment, the myofiber segmentation network uses SegFormer3D.

[0133] It should be noted that after the vertebral body is located through the aforementioned steps, a three-dimensional segmentation mask of muscle and fat in the target area is obtained through manual annotation. Then, the muscle and fat segmentation network is trained using this three-dimensional segmentation mask of muscle and fat.

[0134] Specifically, the training of the myolipin segmentation network uses conventional techniques in the field, so it will not be described in detail here.

[0135] In one embodiment, after obtaining the vertebral body level positioning mask through the above step S5, the human vertebral body level myofiber segmentation method further includes: recording the vertebral body type of the target area, the start and end slice indices of the Z-axis corresponding to the target area, and calculating its center slice index as a representative Z coordinate.

[0136] In one embodiment, after the step of segmenting the muscle and fat in the target area, the human vertebral body level muscle and fat segmentation method further includes: obtaining target parameters corresponding to the target area.

[0137] In this embodiment, the target parameters include, but are not limited to, any one or a combination of two or more of the following: average HU, total voxel volume, muscle cross-sectional area, and muscle index.

[0138] HU, short for Housfield Unit, is a unit of measurement for CT plain scan imaging data, reflecting the density differences of tissues in different parts of the human body relative to water; the average HU can be intuitively obtained from the target area (displayed in 3D CT medical images).

[0139] Voxels are the basic elements of CT plain scan imaging. They are the smallest imaging units for three-dimensional spatial imaging of the human body. Each voxel has a unit volume. The volume of a unit voxel is obtained by multiplying the imaging resolution in the XYZ directions. The total volume of voxels is: the total number of voxels in a set × the volume of a unit voxel. Specifically, in this embodiment, the total volume of voxels can be the total volume of muscle voxels or the total volume of fat voxels.

[0140] The cross-sectional area of ​​a muscle is equal to the ratio of the total volume of the muscle's voxels to the length of the z-axis per voxel.

[0141] It should be noted that the unit voxel Z-axis length is the standardized Z-axis layer thickness mentioned above (corresponding to the preset physical resolution in the above embodiment).

[0142] Muscle mass index equals muscle cross-sectional area / height 2 (" / " represents the division symbol).

[0143] It should be noted that, by acquiring target parameters in this invention, it is beneficial to detect diseases or predict risks in clinical settings.

[0144] In one embodiment, after obtaining the target parameters, the human vertebral body level myofiber segmentation method further includes: checking the rationality of the target parameters.

[0145] For example, in one embodiment, it is determined whether the cross-sectional area of ​​the muscle is greater than or equal to a preset area threshold, and an alarm indication is issued when the determination result is negative.

[0146] In one embodiment, the preset area threshold is set to 150cm². 2 .

[0147] It should be noted that this invention achieves precise vertebral body localization without relying on layer annotation, and reliably incorporates knowledge from publicly available datasets, reducing the computational complexity of 3D segmentation while maintaining localization accuracy, and suppressing layer localization and myolipin composition analysis results (corresponding to the aforementioned target parameters) that do not conform to prior knowledge from neural network output.

[0148] The scope of protection of the human vertebral body level myofiber segmentation method described in this embodiment is not limited to the execution order of the steps listed in this embodiment. Any solution implemented by adding, subtracting, or replacing steps in the prior art based on the principle of this invention is included within the scope of protection of this invention.

[0149] This invention also provides an electronic device, which includes a processor and a memory; the memory is used to store a computer program; the processor is used to execute the computer program stored in the memory, so that the electronic device performs the above-described human vertebral body layer myofiber segmentation method.

[0150] This invention also provides a computer-readable storage medium storing a computer program that, when executed by an electronic device, implements the above-described method for myofiber segmentation at the human vertebral level.

[0151] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing a processor. The program can be stored in a computer-readable storage medium, which is a non-transitory medium, such as random access memory, read-only memory, flash memory, hard disk, solid-state drive, magnetic tape, floppy disk, optical disk, and any combination thereof. The storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. This available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state drive (SSD)).

[0152] This invention also provides a human vertebral body level myofiber segmentation system, which can implement the human vertebral body level myofiber segmentation method described in this invention. However, the implementation device of the human vertebral body level myofiber segmentation method described in this invention includes, but is not limited to, the structure of the human vertebral body level myofiber segmentation system listed in this embodiment. All structural modifications and substitutions of the prior art made according to the principles of this invention are included within the protection scope of this invention.

[0153] like Figure 6 As shown, in one embodiment, the present invention provides a human vertebral body level myofiber segmentation system, the human vertebral body level myofiber segmentation system comprising: The image acquisition module 61 is used to acquire three-dimensional CT medical images of the human vertebrae.

[0154] The resampling module 62 is used to resample the three-dimensional CT medical images to obtain three-dimensional resampled images.

[0155] The mask generation module 63 is used to generate a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image.

[0156] The sequence generation module 64 is used to statistically analyze the pixel ratio of all vertebrae categories contained in the semantic segmentation 3D mask along the Z-axis, slice by slice, so as to generate a one-dimensional dominant vertebrae label sequence based on the pixel ratio of the vertebrae categories.

[0157] Processing module 65 is used to obtain a vertebral level localization mask based on the one-dimensional dominant vertebral label sequence using morphological post-processing.

[0158] The myofiber segmentation module 66 is used to segment the myofiber within the human vertebral body according to the vertebral body level positioning mask.

[0159] It should be noted that the structure and principle of the image acquisition module 61, the resampling module 62, the mask generation module 63, the sequence generation module 64, the processing module 65, and the myofiber segmentation module 66 correspond one-to-one with the steps (steps S1 to S6) in the above-mentioned human vertebral body level myofiber segmentation method. The specific working principle can also be referred to the description of the human vertebral body level myofiber segmentation method in the foregoing embodiments, so it will not be repeated here.

[0160] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, or methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules / units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or modules or units may be electrical, mechanical, or other forms.

[0161] The modules / units described as separate components may or may not be physically separate. The components shown as modules / units may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules / units can be selected to achieve the objectives of the embodiments of the present invention, depending on actual needs. For example, the functional modules / units in the various embodiments of the present invention may be integrated into one processing module, or each module / unit may exist physically separately, or two or more modules / units may be integrated into one module / unit.

[0162] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0163] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0164] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for segmenting myofibrils at the vertebral level in the human body, characterized in that, The method for segmenting human vertebral body myofibrils includes: Acquire three-dimensional CT medical images of the human vertebral body; The three-dimensional CT medical images are resampled to obtain three-dimensional resampled images; Generate a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image; Along the Z-axis of the semantic segmentation 3D mask, the pixel percentage of each cone category contained in the semantic segmentation 3D mask is statistically analyzed slice by slice, so as to generate a one-dimensional dominant cone label sequence based on the pixel percentage of each cone category. Based on the one-dimensional dominant vertebral body label sequence, morphological post-processing is used to obtain the vertebral body level localization mask; The myofibril within the human vertebral body is segmented based on the vertebral body level positioning mask.

2. The method for myofiber segmentation at the vertebral level according to claim 1, characterized in that, The step of resampling the three-dimensional CT medical images to obtain three-dimensional resampled images includes: In order to standardize the Z-axis slice thickness of the three-dimensional CT medical image, voxel resampling is performed using cubic spline interpolation in the Z-axis direction of the three-dimensional CT medical image, thereby obtaining a standardized image. Voxel resampling is performed on the XY plane of the standardized image using linear interpolation to unify the number of pixels in the XY plane matrix of the standardized image, thereby obtaining the three-dimensional resampled image.

3. The method for myofiber segmentation at the vertebral body level according to claim 1, characterized in that, The semantic segmentation 3D mask contains at least one slice along the Z-axis; the step of statistically analyzing the pixel percentage of each vertebral category included in the semantic segmentation 3D mask along the Z-axis, slice by slice, to generate a one-dimensional dominant vertebral label sequence based on the pixel percentage of each vertebral category includes: For each slice, the number of pixels of all cone types contained in the slice is counted. Calculate the percentage of pixels of each vertebral body type in the slice based on the number of pixels; Determine whether there exists a vertebral body category whose pixel percentage exceeds a preset percentage among all vertebral body categories included in the slice; the preset percentage is greater than or equal to 50%. When the determination result is yes, the slice is marked as the dominant vertebra; the dominant vertebra is the vertebra category whose pixel percentage among all vertebra categories contained in the slice exceeds the preset percentage; If the result is negative, the slice is marked as background; The one-dimensional dominant vertebrae label sequence is generated based on the labeling results; wherein, the one-dimensional dominant vertebrae label sequence is a one-dimensional array with a length equal to the number of Z-axis slices, each element in the one-dimensional dominant vertebrae label sequence corresponds to the background or the dominant vertebra, and the value used to label the background is different from the value used to label different dominant vertebrae; the number of Z-axis slices is the number of all slices contained in the semantic segmentation three-dimensional mask in the Z-axis direction.

4. The method for myofiber segmentation at the vertebral level according to claim 1, characterized in that, Before the step of generating a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image, the human vertebral body level myofiber segmentation method further includes: training a 3D segmentation model using a publicly available 3D vertebral body segmentation dataset to obtain a trained 3D segmentation model. The step of generating a semantic segmentation 3D mask covering all regions of the 3D resampled image includes: generating the semantic segmentation 3D mask based on the 3D resampled image using the trained 3D segmentation model.

5. The method for myofiber segmentation at the vertebral body level according to claim 1, characterized in that, The morphological post-processing used to obtain the vertebral level localization mask based on the one-dimensional dominant vertebral body label sequence includes: The one-dimensional dominant vertebral label sequence is subjected to one-hot encoding to obtain a first sequence; each first sequence corresponds to one vertebral category, and the number of first sequences is equal to the number of vertebral categories; The first sequence is expanded using an expansion operation with a kernel length of a first preset length to fill the mask gaps in the first sequence, thereby obtaining the second sequence. The second sequence is etched using an etch operation with a etch core length of a second preset length to remove isolated noise from the second sequence and obtain the third sequence. By traversing the third sequence, the length of each vertebral body category is determined, and it is determined whether the length meets the corresponding preset length condition. When the length meets the preset length condition and the vertebral body category meets the vertebral body spatial order constraint rule, the third sequence corresponding to the target category is used as the vertebral body level positioning mask of the target category. The target category is the vertebral body category whose length meets the preset length condition and meets the vertebral body spatial order constraint rule.

6. The method for myofiber segmentation at the vertebral body level according to claim 1, characterized in that, The step of segmenting the myofat within the human vertebral body layer according to the vertebral body layer positioning mask includes: The target region is extracted from the three-dimensional CT medical image based on the vertebral body level positioning mask; the target region is the image region in the three-dimensional CT medical image corresponding to the vertebral body category corresponding to the vertebral body level positioning mask. The muscle and fat within the target area are segmented.

7. The method for myofiber segmentation at the vertebral body level according to claim 6, characterized in that, After the step of segmenting the muscle and fat within the target area, the human vertebral body level muscle and fat segmentation method further includes: obtaining target parameters corresponding to the target area; the target parameters include at least one or two or more combinations of the following: average HU, total voxel volume, muscle cross-sectional area, and muscle index.

8. A human vertebral body-level myofiber segmentation system, characterized in that, The human vertebral body level myofibril segmentation system includes: The image acquisition module is used to acquire three-dimensional CT medical images of the human vertebrae; The resampling module is used to resample the three-dimensional CT medical images to obtain three-dimensional resampled images. The mask generation module is used to generate a semantic segmentation 3D mask covering all regions of the 3D resampled image based on the 3D resampled image; The sequence generation module is used to calculate the pixel ratio of all cone categories contained in the semantic segmentation three-dimensional mask along the Z-axis, slice by slice, so as to generate a one-dimensional dominant cone label sequence based on the pixel ratio of the cone categories. The processing module is used to obtain the vertebral body level localization mask based on the one-dimensional dominant vertebral body label sequence using morphological post-processing. The myofiber segmentation module is used to segment the myofiber within the human vertebral body based on the vertebral body level positioning mask.

9. An electronic device, characterized in that, The electronic device includes: a processor and a memory; The memory is used to store computer programs; The processor is used to execute the computer program stored in the memory to cause the electronic device to perform the human vertebral body layer myofiber segmentation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by an electronic device, the program implements the human vertebral body level myofiber segmentation method as described in any one of claims 1 to 7.

Images